Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
ISSN 2282-6483
The Causal Effect of Scholarships Targeted at Low Income Students on
Performance: Evidence from Italy
Veronica Rattini
Quaderni - Working Paper DSE N°968
The Causal Effect of Scholarships Targeted at Low Income
Students on Performance: Evidence from Italy
Veronica Rattini∗
Abstract
This paper exploits discontinuities in the assigment of scholarships targeted at low income
students in an Italian University in order to evaluate the effects of monetary incentives on students’
academic achievement. Results reveal positive and sizeable causal effects both in terms of credits
and grades. Gender differentials also emerge: male students drive the results on credits outcome
while females students drive the effect on grades. These results suggest that the scholarship design
contributes to reducing the probability of delayed graduation (“Fuori Corso” problem).
JEL codes: H5, I21, I28, .
Keywords: Human Capital, Monetary Incentives, Regression Discontinuity
∗Contact author: [email protected]
1
1 Introduction
Given the extensive social and private benefits that result from tertiary education, inclusive ac-
cess and a well functioning system are essential for achieving social justice and enhancing national
human capital. Furthermore, if one considers the strong correlation between tertiary education en-
rollment and family background (McPherson and Schapiro, 2006), concrete and effective initiatives
are necessary to provide better opportunities of access and success for students from lower income
families and minority groups.
In Italy, the “Right to Study” Constitutional principle responds to such equity purposes by
providing different types of services for students from lower income family who want to enroll at an
university program. In this research, I focus on the “Right to Study Scholarship”. In particular,
I analyze how the monetary provision of such scholarship causally impact on students’ academic
achievements by applying a Regression Discontinuity analysis.
I interested in understanding whether these monetary incentives could serve to boost students’
performance by reducing direct and indirect college costs. Moreover, since the maintenance and
non-reimbursement of the mean-tested scholarship are conditional on the achievement of minimum
credit requirements, the analysis gives the opportunity to understand whether such requirements
are adequately defined in order to serve the purpose of enhancing students’ performance and the ex-
tent to which they could alleviate the problem of being “Fuori Corso” (delayed graduation), which
is a wide-spread phenomenon in many countries (see, among others, Hakkinen and Usitalo, 2002;
Van Ours and Ridder, 2002; Brunello and Winter-Ebmer, 2003; Bowen, Chingos and McPherson,
2009).
In the economic literature, a growing number of empirical studies addresses the question of
how monetary incentives shape students’ educational choices. In particular, a number of papers
investigate the effect of such incentives on the choice of school program and on degree completion
(Hansen 1983; Kane 1994; Dynarski 2000,2003; Van der Klaauw 2002; Goodman 2008). Other
works, mostly from US, have focused on the effects on academic performance and do not find uni-
vocal results (Angrist and Levy 2002; Leuven et al. 2003; Cornwell at al. 2003; Dynarski 2005;
2
Belot et al. 2007; Angrist et al. 2009).
For what concerns the case of Italy, the evidence is not uniform and the present study ought
to provide new insights on the issue. In particular, Garibaldi et al. (2012) find that an increase in
college cost, in response to delayed graduation, exerts a remarkable effect on the on-time comple-
tion rate, using evidence from Bocconi University. In line with this evidence, I find that monetary
incentives speed up the number of credits a student achieves in the first year of the degree without
prejudice to, but actually increasing, average grades. This effect can arguably help the “Fuori
Corso” problem. In other words, the need-based scholarship leads low-income students to exert
more efforts in order to meet minimum credit requirements and to avoid losing their grants (which
is equivalent to an increase in university costs).
De Paola et. al (2010) find that “financial rewards increase student performance both in terms
of number of credits acquired and grades obtained at exams”. My results are in line with such
evidence, even if their institutional context is different with respect the one analyzed here. In
fact, they analyzed the effect of an extra reward financed through the European Social Found for
above-average students, which could not give the same incentives as the Right to Study Scholarship.
Mealli and Rampichini (2012) have, instead, analyzed data regarding four Italian universities
and they show that the “Right to Study Scholarship” (which is the same Italian institution that
we here analyze) prevents students from droping-out. However, they find that the effect vanishes
for lower income students. Here we provide new evidence from the “Right to Study Scholarship”
at the University of Bologna showing that such effects on performance for lower income students
are not only significant but are also remarkably large.
My results are also at odds with Schizzerotto et al. (2012) who studies the effect of a merit-
based scholarship at the University of Trento and do not find any significant effect on academic
achievement. Such contrast might have to do with the fact that they use a relatively small sample
from survey data, which are more sensible to measurement errors. In my case, the administrative
dataset I use allows to gain precision on the estimates.
3
The paper is organized as follow. The next section outlines how the present study relates to the
literature developed so far. Section 3 analyzes the institutional framework and Section 4 describes
the data and methodology. In section 5, I will present and discuss results on the causal effects of
scholarships on different measures of academic achievement. Section 6 discusses the robustness of
the findings and Section 7 concludes.
2 Literature Review
In recent years a growing number of interventions designed to increase tertiary education enrol-
ment, completion and effectiveness took place. A number of these programs targeted low income
students. While most of the interventions are based on monetary incentives, program design fairly
varies. Examples of the designs are: merit and need-based scholarships, tuition subsidies or part-
time working programs.
Such policies are justified by the fact that large economic returns related to tertiary educa-
tion have been clearly documented and they do not constitute only private benefits (Barro 1991,
Acemoglu, D., and J. Angrist 1999, Temple 2000, Dasgputa 2001, Glaeser et al. 2004). Positive
externalities resulting from education provide a strong economic reasoning for such interventions.
Moreover, it has been shown that the effort provided by students in schooling activities is one of
the most important factors in the human capital accumulation, possibly more effective than teacher
quality and school resources (Costrell, 1994; Bonesronning, 2004).
Further justifications come from the fact that, in economic theory, monetary aids are supposed
to act as incentives promoting effort and scholastic performance (Lazear 2000) being, therefore,
helpful in reducing the still nowadays large percentage of young students who face severe problems
in educational choice. It is well documented, in fact, that the share of students who drops-out
the tertiary education is commonly high even in developed countries and that time to degree com-
pletion has increased remarkably in the last two decades. 1 Therefore, the interest in how such
1For example, Bound et al. (2006) for US and Brunello and Winter-Ebmer (2003) for Europe
4
aid programs are related to students’ performance started to grow in economic research. However,
there seems to be much room from further research since results are by no means univocal: some
studies provide evidence of a positive impact, while others find rather weak effects. This hints at
the necessity of better understanding of how different policy designs implicate in different effects
on performance and how these effects vary across different groups of subjects.
In non-experimental research the effectiveness of academic support services is mixed and do
not provide a realiable causal measure of such relationship (e.g., surveys by Bailey et al. 2005;
Pascarella et al. 1991; Lotkowski et al. 2004, and Wyckoff et al. 1998). On the other hand, better
designed studies using experimental and quasi-experimental designs draw a more clear picture.
With reference to the US case, Bettinger (2004) shows that a mean-tested programme of financial
assistance exerts a remarkable reduction in the drop-out rate in Ohio University. Dynarski (2005),
who analyzed the effect of the Georgia’s Helping Outstanding Pupils Educationally (HOPE) merit-
based program, finds a positive effect in college completion by 3 to 4 percentage points. Cornwell at
al. (2003) analyze the same program and find that the shift from need- to merit-based aid increases
the probability to withdraw and reduces the number of average credits obtained. Angrist and Lavy
(2009) evaluate the effectiveness of financial rewards on the academic achievement of Israeli stu-
dents. They find that the program has led to relevant effects for female but not for male. Differences
among genders emerge in Angrist, Lang and Oreopoulos (2009), who conduce an experimental eval-
uation on students enrolled to a large Canadian University. The results, in fact, show that while
male students were not affected by the program, female ones improved their academic performance.
For what concerns Europe, Leuven et al. (2003) analyze the effect of merit-based rewards
for students enrolled at the University of Amsterdam. They find that only high-ability students
and those with highly educated fathers obtain better academic achievement when assigned to the
high reward group. For the whole sample taken together instead, there was no effect. Belot et
al. (2007) exploit a Dutch reform in the educational system in order to understand the effect of
student assistance on academic performance and students’ time allocation. They find that there is
a small positive effect on the grades but drop-out and time allocation of students remain basically
5
unchanged. De Paola et al. (2010) analyze the effect of a randomly assigned merit-reward for
students enrolled at University of Calabria. They find positive effects of the monetary incentives
on average grades. Garibaldi et al. (2012) find that at Bocconi University in Milan, the increase
in the tuition payments for “Fuori Corso” students, i.e their enrollment in the university system
extended beyond the legal length of their program, exert a remarkable effect on on-time completion
rate.
Mealli and Rampichini (2012) analyze data regarding four Italian universities and show how public
provided university scholarships, under the “Right to Study Scholarship”, prevent drop-out for those
students from medium income families, while for poorer students the effects are not significant.
Schizzerotto et al. (2012) looks instead at the incentives assigned to students from low income
families introduced in a well-defined area in the North-East of Italy, the administrative province of
Trento. They find no effect whatsoever on the drop-out rate, the average mark and the number of
credits achieved.
In sum, from a broad review of the recent empirical literature, it becomes clear that the overall
evidence on the effect of monetary incentives is rather controversial, ranging from studies finding
no effects whatsoever to studies finding large and positive effects on performance.
The goal of this paper is to enrich this debate showing the results of a mean-tested program
carried out in the Italian context. In particular, I try to disentangle the pure effect of the so called
“Right to Study Scholarship” from other confounding factors on students’ academic achievement
through applying a Regression Discontinuity Design. The idea underlying the “Right to Study-
Scholarship” is to reduce the cost related to tertiary education attendance in order to prevent low
family income students from being excluded, drop-out, or under-perform with respect to other
pupils due to financial constraints. This study show how such institution could be used to serve
the additional pourposes of reducing the percentage of “Fuori Corso” students and, more generally,
in fostering the accumulation of human capital.
6
3 The Institutional Framework
Tertiary education in Italy is generally accessible to students with an high school diploma, indepen-
dently of the type obtained (scientific, classical, professional). Moreover, the Italian Constitution
states in art. 34 that University education should be accessible also to those students who do not
have the financial resources and the that the Italian Republic furthers the realization of the, so
called, “Right to Study” by providing scholarship and other allowances to students.
“Pupils of ability and merit, even if lacking financial resources, have the right to attain
the highest grades of studies. The republic furthers the realization of this right by schol-
arships, allowances to families, and other provisions, to be assigned through competitive
examinations.”
To enhance equal opportunity and fair access, the region offers many types of services: al-
lowances for international mobility, housing and meals services, services for people with disability,
vouchers for education programs (Master, High-level education, etc.), fiduciary loans, and part-
time working possibilities. The present study will be focused on the “Right to Study Scholarships”
offered for the University of Bologna. The “Right to study Scholarship”, from now on “RSS”, is
one of the services provided by the Region of Emilia-Romagna, where the University of Bologna
is located, to promote access and equity in education. In particular this research want to measure
what are the effects of such monetary incentives on students’ academic achievement.
The “Right to Study” service in Emilia-Romagna has generally a large coverage, both in terms
of the number of recipients and of the amount of financial resources provided. Specifically, 13.475
students received a RSS over a total of 98.357 students in the region (13,7%) for the 2008/2009
academic year. 2 Furthermore, to finance the RSS for the same academic year, the Region of Emilia-
Romagna collected: from national resources e151.986.000, from regional resources e158.120.201 ,
from regional taxes e171.085.441, making a total of e481.191.642. 3
In this perspective, it becomes quite important to know how these grants shape students’ in-
centives and, in particular, whether they have any effects on students’ academic achievements,
2Data from “Ministry of Education, University and Research - MIUR”3Data from “Ministry of Education, University and Research - MIUR”
7
especially in Italy where students’ profiles and performance are quite peculiar. In fact, it has
been estimated 4 that 36% of students are “Fuori Corso”, i.e those whose their enrollment in the
university system extended beyond the legal length of their program, and 21,7% of students drop-
out. Moreover, the OECD 5 estimates that in Italy 23,2% of young in the 15-29 age bracket are
NEET-Not in Education, Employment, or Training and that the employment rates of the youths
in 25-29 age bracket is one of the smallest in Europe. In this context, therefore, proper monetary
incentive schemes could play a role not only in providing fair access to the University but also in
fostering students’ performances and enriching national human capital, as through the prevention
of university drop-out and the reduction of the share of “Fuori Corso” students.
The regional agency appointed for the distribution of the “Right to Study” services is ER.GO
and from 2008 onward, the agency have fully covered all the scholarship’ applicants, a 100% suc-
cessful rate. The application to scholarships is made before the academic year starts. 6 Results on
allowance eligibility are published within few months and the first installment of the grant (50% of
the yearly allowance) is paid due the end of calendar year. The second half of the financial transfer
is bind to the satisfaction of precise credit requirements, which are known ex-ante.
The scholarships differ according to the student status which depends on their place of residence:
“In sede”, i.e their university’s course is in the city of residence or they do not live more than 45
minutes far (by public transport) from the university’s center; “Fuori sede”, i.e their university’s
course is more than 90 minutes far (by public transport) away from the city of residence, “Pen-
dolari”, i.e their university’s course is between 45 and 90 minutes away from the city of residence.
There is no distinction, on the other hand, with respect to the level of their degree. Within groups,
“In Sede”, “Fuori Sede”, “Pendolari”, scholarships are assigned according to three thresholds on
one income indicator of the family, ISEE. Furthermore, eligibility is always conditional on a max-
imum value of a further wealth indicator of the family, ISPE, which shall not exceed e32.320,64.
The ISEE indicator is given by the annual after-tax income plus the 20% of family assets, adjusted
for the family size by means of an equivalence scale; the ISPE is instead an indicator based just
4AlmaLaurea – Annual Report on University’ Graduates 20135Education at la glance 20136At the University of Bologna all the courses start from September/ October to June/July of each year.
8
on the family assets and it is adjusted for the family size by means of an equivalence scale. The
amount of scholarships are the following:
Table 1: Scholarship Value
ISEE Threshold “Fuori Sede” “Pendolari” “In Sede”
Until e 11,927.35 e 5,073.78 e 3,043.88 e 2,255.11
From e11,927.36 to e 14,909.19 e 3,942.83 e 2,420.89 e 1,828.83
From e 14,909.20 to e 17,891.03 e 2,811.88 e 1,796.93 e 1,402.53
For the undergraduate freshman, scholarship’s assignment is conducted just on the basis of the
two economic indicators, ISEE and ISPE. The master freshman, should have obtain, in addition,
at least 150 credits in the undergraduate program. On the top of that, the grant maintenance
and right not to reimburse the scholarship are conditioned to precise credit requirements. Namely,
undergraduate students should obtain 25 credits (out of 60) by the end of the first academic year
and master students should obtain 30 credits (out of 60) by the end of the first academic year,
independently of their status, “Fuori Sede”, “In sede”, “Pendolari”.
For second year applicants, the assignment is conducted on the basis of the two economic indi-
cators, ISEE and ISPE, and on the basis of the credits obtained in the first year of study. Precisely,
in order to be eligible for the scholarships, students should have obtained at least 25 credits in
the first year of undergraduate course or 30 credits if enrolled in the first year of a master course.
Moreover, in order not to reimburse the grant, recipients should obtain at least 80 credits at the
end of the second year of enrolment (both for undergraduate and master program). The same
credit requirement (80 credits by the end of the second year) is used for the third year application,
together with the two economic indicators. In addition, students enrolled to the third year of the
undergraduate program should obtain at the end of the academic year 135 credits not to refund
ER.GO for the scholarship amount.
On the top of credit requirements, there are also “credits bonus”. The design of the RSS offers
the possibility to students to use credits bonus, ie. they can add a limited numbers of credit bonus
9
Table 2: Credits Requirements by the end of each year
Type of course 1st Year 2st Year 3rd Year
Undergraduate 25 80 135
Master 30 80 -
to the number of actual completed credits, in order to meet the requirements. Specifically, students
have the right to use:
a) 5 credits bonus if they asked to satisfy the 2nd year requirement;
b) 12 credits bonus if they asked to satisfy the 3rd year requirement;
Recipients can take advantage of using current year remaining credit bonus in further years,
however these are not cumulative.
4 Data and Methodology
The data are provided by the regional agency which is in charge of delivering the RSS, ER.GO,
and by the University of Bologna. The advantage to use administrative data is that observations
on performance and income status are more precise than survey data, therefore it allows for more
precise estimates and decreases the risk of measurement errors biases.
The academic year of reference is the 2008/2009 and the sample includes all the freshman admit-
ted in all the twenty three faculties of the University of Bologna. 7 The data set includes students
enrolled at different levels of Tertiary education: Undergraduate level and Master level. The data
set also includes information on students’ demographic characterstics and on university-related
variables: high-school grade, credits and average mark obtained, university center (Bologna, Forlı,
Cesena, Rimini, Ravenna), faculty, graduation course, type of scholarship obtained (“In Sede”,
“Fuori Sede”, Pendolari”), ISEE and ISPE.
7From academic year 2012/2013 the University of Bologna had change the organisation from 23 faculties to 11 Schoolsand 33 Departments.
10
Table 3 in the appendix provides descriptive statistics for the sample of students. About 60%
percent of the students are female. The average age is 21 years old. The geographical distribution is
quite symmetric as 37% of students are coming from the North, 26% from the Center, 28% from the
South of Italy and 9% are foreign students. The High School grade in Italy ranges from 60 to 100;
the sample mean equals 81.6 points. The high school grade is set to missing for foreign students
since I do not have enough information to perfectly convert the foreign scale into the Italian ones.
The average number of credits obtained during the first year of enrollment is around 31 with an
average grade of 26. Some data on average grade were missing in the data provided by ER.GO and
could not be recovered until the time being. “In Sede” and “Fuori Sede” scholarships’ students are
quite evenly distributed in the sample.
In order to study the effect of scholarships on students’ performances, I use a Regression Dis-
continuity Design. The RDD has been largely used in economics and behavioral sciences and was
firstly introduced by Thistlethwaite and Campbell (1960). The attractiveness of such design is that
it allows to identify and estimate treatment effects in a context similar to a formal randomized ex-
periment. In particular, data on random assignment of treatment and control groups are commonly
used in medical analysis. In economic research and, precisely, in the evaluation of social programs,
the use of random assignment is mostly controversial and remains unusual. Under this perspective,
RDD succeeds in establishing soft conditions that allow to identify and estimate treatment effect
on observational studies, without relying on parametric evaluation model, often criticized in the
literature (for example Lalonde(1986)).
The identification and estimation issues for the RDD were formalized in the work of Hahn,
Todd, and van der Klaauw (2001), which describes the minimum set of conditions under which it
possible to nonparametrically identify the treatment effect.
It is typically imagined that, for each individual i, there exists two potential outcomes: Zi(1)
if the unit is exposed to the treatment and Zi(0) if not exposed. The difference Zi(1) − Zi(0)
represent the causal effect of the treatment. The real problem of causal inference is that we cannot
observe the pair Zi(0) and Zi(1) simultaneously (Holland (1986)). We therefore typically focus on
average effects of the treatment, E[Zi(1)|Y ] and E[Zi(0)|Y ] over (sub-)populations, rather than at
11
individual level. The idea of the quasi-experimental RD design is to exploit the discontinuities in
the treatment assignment. Precisely, assuming that units close enough to the threshold are similar
except for their treatment status, the observed difference in outcomes can be attributed just to the
treatment. Formally, consider that all individuals to one side of the a certain threshold, Y = c,
are exposed to treatment and all those to other side are denied treatment. Therefore, we only
observe E[Zi(1)|Y ] that is the average treatment effect for the treated units and E[Zi(0)|Y ] for the
untreated ones. By the continuity assumption is possible to show that we can estimate the average
treatment effect as follow:
ATE = limε→0
E[Zi|Yi = c+ ε]− E[Zi|Yi = c− ε] = E[Zi(1)− Zi(0)|Yi = c] (1)
In our setting, the set of outcome variables, Zi, consists of: number of credits and average mark
obtained in the first year on enrolment, a constructed index of performance 8 and by a binary vari-
able taking values [0,1] if the student have satisfied the credits’ requirements. The treatment is a
dummy variable taking value 1 if the individual receives the higher scholarship (around e840 more
with respect to the counterfactual) and 0 otherwise; the treatment rule is determined by ISEE’s
thresholds. For the first threshold, the rule is the following: students with the income indicator,
ISEE, lower or equal to e11.927,35 are assigned to the higher scholarship, while students with
ISEE between e11.927,36 and e14.909,19 are assigned to the lower one; the difference between
the two scholarships is of about e950. For the second threshold, students with an income between
e11.927,36 and e14.909,19 receive a scholarship of e730 higher than the students with ISEE be-
tween e14.909,20 and e17.891,03. The mean aggregate difference of e840 is computed giving equal
weight to the change at the two thresholds.
In order to aggregate students with ISEE near the two thresholds, ci = (c1, c2), I constructed a
8The index is constructed by firstly normalizing the credits and the average obtained between [0;1] and then bycomputing the weighted average of the two measures, giving two equal weights.
12
new variable Yi that is define by:
Yi =
0 if ISEE = ci
+ (ISEE − ci) if ci ≤ ISEE ≤ ci +ci+1 − ci
2
− (ISEE − ci) if ci −ci+1 − ci
2≤ ISEE ≤ ci
(2)
Notice that the positive (negative) values of Yi are computed taking just the values of ISEE
that are greater (smaller) than the threshold values plus the first half of the consecutive threshold’
bracket. In this way we are not confounding the scholarships’ incentives of students near the first
threshold with those of students near the second threshold, and the ones of students near the second
threshold with those near the third threshold.
From the treatment analysis on the two thresholds, I will estimate the Local Average Treatment
Effect (LATE). In particular:
Zi = α+ F (Yi) + γDi + εi (3)
where the Zi s a vector of outcomes variable, F (Yi) is a polynomial of 4th degree on the constructed
variable defined as Yi, Di is the dummy variable taking value equal to 1 if students are assigned to
the higher scholarship. Then it is possible to demonstrate that:
γ = E[Z|Y = 0+]− E[Z|Y = 0−]
is an unbiased estimator of the LATE.
5 Results
The results are drawn combining data of students who have a value of ISEE near the first and the
second scholarships’ thresholds (e11.927,35 and e14.909,19 respectively). By aggregating those
data, the analysis gains in term of precision but we lose information on how the effects of interest
vary with the ISEE and with the students’ status (“In Sede”, Fuori Sede”, “Pendolari”). The
results are presented in Table 4.
13
The local average treatment effects are estimated allowing for different polynomials on the two
sides of the aggregated threshold. To check for the robustness of the results, I use four specifications
with polynomials of different degrees, from the 1st to the 4th degree. As standars in such method-
ology, when I increase the order of the polynomial form, I simultaneously enlarge the observational
range in order to avoid losing the precision of estimates. More precisely, for the 1st polynomial
degree I used a bandwidth of ± e500 around the aggregate ISEE threshold. For the 2nd degree,
I add e350 obtaining a range of ± e850 while for the 3rd degree we use those observation lying
around ± e1200 with respect the aggregate threshold. For the 4th degree we use a bandwidth of
± e1400, the maximum possible range (half of the distance between the first and second threshold).
As it is shown in table 4, students who receive an higher scholarship amount, on average e840
more, have better results on academic achievement during the first year of university enrollment.
In particular, according to the specification used, students with the higher scholarship acquire from
9 to 18 credits more with respect to students with a lower scholarship during the frist year. Con-
sidering that the regular number of credits for the first year is 60 and that the average number of
credits obtained in the sample is around 32, the size of the effect is arguably large (from 28% to 56%).
The average of first year grades of those students with higher scholarships is around 2 points
higher than the one of those with a lower financing. The grades scale in Italy ranges from 18
(sufficiency grade) to 30 points (plus laude). Since the sample mean of the average first year grade
is of about 26 points, the estimated causal effect of an e840 higher scholarship is of about the 7%
on the average mark.
It is important to notice that, fellowship need not be reimboursed if the credit requirements is
achieved. Students aiming at fullfilling this requirement only may pass more exams but with lower
grades. I check if this is the case by looking at the effect on students’ GPA at the end of the year.
Pure credit requirements seem to be sufficient to increase students’ performance not only in terms
of credit achievement but also with respect to marks. In fact, as shown in the Panel C, an index of
performance (made out of an average on normalized credit achievement and average marks) jumps
of about 0.1 decimal points for students with the higher scholarship.
14
Finally, I analyze the effect on the probability of having satisfy the credits requirements for
the maintenance of the scholarship, which are set equal to 25 (30) credits for the undergraduate
(master) students. The average rate of success in the sample is of about 78%, a fairly large fraction.
Nevertheless, the results show that the monetary incentives still have a positive effect on the prob-
ability of having satisfied this requirement. In fact, the effect goes from a minimum of 0.22 points
to a maximum of 0.47 probability points, which is equal to an average percentage effect of 28% and
60% respectively. What is clear from the analysis is that the current scholarship design could be
further used as an effective instrument for reducing the percentage of “Fuori Corso” students.
It is interest to contrast this overall finding, that the RSS boosts students’ achievement, with
the results found in Angrist et al. (2009), which instead find no effects on first year grades, of
pure monetary incentives in a randomized experiment. A possible explanation of such divergence
is that in the Canadian contest, the STAR program consists in subsides on tuition, which do not
have the same role of the Right to Study Scholarship. In this sense, the RSS is designed for helping
those students who otherwise would not have the means to attend tertiary education and this is a
decisive incentive. The students analyzed in Angrist et al (2009) are already enrolled at the time
of the experiment in the University.
Moreover, the effect presented above adds a new insight for the Italian empirical literature. Dif-
ferently from Schizzerotto et al. (2012), who have not found any significant effect of mean-tested
fellowships on students’ performances, we show that monetary incentives are useful to enhance aca-
demic achievement. One reason for this discrepancy might be that since their work use survey data
and a relatively small sample to apply a Regression Discontinuity Design, their estimates might not
have been precise enough to allow them to find significant results. In addition, Mealli and Rampi-
chini (2012) have found significant effects of the RSS only on drop-out of students enrolled at the
Universities of Catania, Milan, Padova and Salerno, without focusing on students achievements.
They do not find any significant effect for the lower income students. This is in contrast with the
present results which are focused on lower income students.
I proceed with the analysis in Table 5 by looking at the scholarships’ effects on the probability
15
of obtaining the total number of credits expected for the first year of enrolment at the date of 10
August 2009, which we take as an indicator directly related to the probability to be “Fuori Corso”.
9 Although these results have a positive sign (around 0.1 points) and seem to be fairly stable
over the different polynomial specifications, they are not significant. Given that the analysis above
suggets that the minimum credit requirements may have significant effects on performance (credits,
grades and probability of meeting the requirement itself), a possible experimentation for policy
would be making such minimum requirements closer to the regular number of credits expected
from students along the years (60 credits per year). This could allow the RSS to more effectively
serve the purpose of reducing the share of “Fuori Corso” students.
In order to gain more insight on the underlying relationship between the monetary incentives
and students performance, I analyze treatments effects on the two thresholds separately. The aver-
age increase on the scholarship amount at the first threshold is of around e950 while the difference
at the second threshold is mildly smaller, around e730. The estimates are reported in Tables 6 and
7.
For the first threshold, I find positive and significant effects for all the outcome variables. For
the second threshold instead, results are less clear cut. Treatment effects are only significant on
the number of credits and on the probability of having satisfied the maintenance requirements.
Considering that the sample size and estimated standard errors are similar around both thresholds,
such results suggest that the scholarship is more effective in boosting academic performance of
lower income students. Even if we consider that the average increase in grants around the second
threshold is about 24% lower, the decrease in the size of the effect on the number of credits achieved
and average grade is fairly lower than that. This interpretation suggests that policy makers could
allocate a higher share of funds to the less well-off recipients if the goal is to increase the number
of credits and grades.
Finally, I investigate how treatment effects change within sub-groups of population. More specif-
ically, I firstly analyze the results by students status, “Fuori Sede”, “In sede” and “Pendolari, and,
9“Fuori Corso” status is assigned when the student do not acquire 60 credits at the 31/03/2010.
16
unfortunately, I do not find any significant and consistent result for our different specifications. The
same happens when differentiating for Undergraduate and Master students. This may suggests that
the sample is not large enough to allow us to obtain precise enough estimates on these subgroups.
However, I was able to successfully exploit differences in effects between female and male stu-
dents. Results are reported in the Table 8 and 9, for female and male respectively. I find that the
large and significant scholarship effects on credit achievement are mostly driven by male students.
Although still positive, these effects for female are smaller and not significant. On the other hand,
the previous aggregate results on average grades are mainly driven by female students and they
are still positive but not significant for male. These results show that monetary incentives are
differently perceived between male and female students, as it is found in previous literature. In
the present research, I find pure monetary incentives to be effective in raising the academic per-
formance of both genders, but in a heterogeneous way. For male, the significant effect is on credit
achievement, while for female it is on grades.
6 Robustness
In order to further assess the validity of our results, I perform a standard graphical analysis both
on treatment and pre-treatment variables. Results are reported in Figures 1 and 2. I allow for
a 4th order polynomial on both side of the threshold and show a 90% confidence interval around
estimates. As it is possible to see, the graphical analysis confirms the results found with regressions
estimates. For all the treatment variables (except for the average for which results is less clear cut),
we find a significant and positive jump before the aggregate threshold, meaning that those students
just before the threshold, i.e with an higher scholarship, perform better than those just after it, i.e
with a lower scholarship.
Since it is impossible to test for the continuity assumption directly, I can test some implications
of it. More specifically, all the pre-treatment characteristics should have identical distribution, at
the limit, on the two sides of the cutoff. Therefore, in order to test for the validity of our identifica-
tion strategy, I test for a discontinuity in all of our pre-treatment variables applying the same four
17
specifications used for the post-treatment outcomes. I find no significant discontinuity on these in
any of these covariates, which is reassuring for the validity of our identification strategy. These
results are reported in Table 10.
In Figure 2 I show instead a graphical analysis on these pre-treatment covariates: Sex, Age,
High-School Grade, ISPE, Probability to be “Fuori Sede” students. Once more, I find no significant
jump in any of the pretreatment variables. Average High-School Grade seems to be higher for those
students with the higher scholarship, but, as shown in the regression estimates, this difference is
not significant. Moreover, I test further for this possible ability bias by looking at the probability
to be “Fuori Sede”. In fact, if students with higher High-School Grade, decide to go to college only
if they get higher scholarships, I should see a jump in the probability to be “Fuori Sede” since these
are those getting higher grants. Estimates show that this is not the case.
A further important issue to the validity of the identification strategy has to do with manip-
ulation of the assigned variable. Since the value of scholarships are based on the tax report of
family income (ISEE), there could exist some problems of manipulation. Fortunately, I can test
for this hypothesis. First, I test for the continuity of the ISPE (wealth indicator) variable at the
threshold, as for other covariates. Second, I perform the McCrary test (2008) on the density of
assignment variables around the thresholds. Results are available in Figure 2 and 3. As shown
in the Figure 3, there is no significant discontinuity in the density of the observations around the
threshold, supporting therefore the hypothesis of no manipulation on the assignment variable, ISEE.
Concluding, after performing all these robustness tests, we should be confident on the validity
of the identification strategy and on the estimates of the Local Average Treatment Effects of the
scholarship on students’ academic performance.
7 Conclusions
This paper has studied the role of monetary incentives in an educational context. In economic theoy,
monetary incentives are the standard tool to shape indiviual actions and effort and, nowadays,
18
many policy interventions are designed in order to use this principle to influence individual decision
making. Recently, more and more attention is dedicated on how financial incentives could serve
also to condition students behavior in their academic studies. The present paper, in particular, is
focused on a public insistution designed principally to provide fair access and equal opportunity
in higher education. As it emerges from the results of this research, the scholarships provided for
low-income students, could not only respond to the equity purpose cited above, but they could
also be used as active tools to boost students performances. In particular, I found that on average
e840 more will bring students to get from 9 to 18 credits more by the end of the year, with gains
also for their GPA, i.e around two points more (the 7% of the average mark). Moreover, I found
that the scholarships are more effective in this purpose for lower income students (first threshols
results), suggesting that policy maker who wants to magnify such effect should allocate more and/or
higher scholarship for this type of students. In line with the literature, I also find gender gaps. In
particular, female students drive the results on the GPA (probably because they acquire already
an above average amount of credits) while male students drive the finding on credits. The results
presented seem to be valid and robust to different consistency checks as I shown in the robustness
section.
With the present paper, I want to add new causal evidence on how financial incentives shape
individual actions, particularly in an educational context. I find that pure monetary incentives can
significantly increase academic performance in a context of lower income students. Moreover, the
aim of this research is also to provide recommendations for policy maker on how to properly design
instutions that want to effectively boost academic perfromances. Further research should explain
through which channel the financial incentives are working. In particular, is suggested to acquire
additional information in order to understand if students are studying better just because of the
threath of loosing the scholarship and/or because they have a positive shock on their income, with
consequences on their labor supply. Responding to such questions, will allow to better target the
scholarship and the related requirements.
19
References
[1] Angrist, Joshua D. and Victor Lavy (1999) “Using Maimonides rule to estimate the effect of
class size on scholastic achievment”,Quarterly Journal of Economics, May, pp. 533-575.
[2] Angrist, Joshua D. and Victor Lavy (2002), “ The Effect of High School Matriculation Awards:
Evidence from Randomized Trials”, NBER Working Paper No. 9389.
[3] Angrist, Joshua D., Daniel Lang and Philip Oreopoulos (2009), “ Incentives and services for
college achievement: evidence from a randomized trial” ”,American Economic Journal, Vol.1,
No.1, pp 136-63.
[4] Bertola, Giuseppe and Daniele Checchi (2003),“Education financing and student achieve-
ment”, Swiss Review of Education - Revue suisse des sciences de l’education, Vol. 25, No. 3,
pp 431-54.
[5] Belot, Michele, Erik Canton, and Dinand Webbink (2007) “Does reducing student support
affect scholastic performance? Evidence from a Dutch reform”. Empirical Economics, Vol.
32 No 2-3 pp. 261–275.
[6] Bound, John, Michael Lovenheim and Sarah Turner (2006), ”Understanding the Increased
Time to the Baccalaureate Degree”, Mimeo, Univesrsity of Michigan.
[7] Bowen, W.G., Chingos M.M. and McPherson M.S. (2009), Crossing the Finish Line: Com-
pleting College atAmerica’s Public Universities, Princeton University Press, Princeton.
[8] Brunello, G. and Rudolf Winter-Ebmer (2003), “Why do students expect to stay longer in
college? Evidence from Europe”, Economics Letters, Vol. 80, No. 2, pp. 247-253.
[9] Brunello, G. and Checchi D. (2007), “Does School Tracking Affect Equality of Opportunity?
New International Evidence”, Economic Policy, 22(52): 783-861.
[10] Checchi, Daniele, (2002), “Formazione e Percorsi Lavorativi dei laureati dell’ Universita degli
Studi di Milano”, Working Paper n. 14.2002, Universit‘a Statale of Milan.
[11] Cornwell, Christopher, Kyung Hee Lee, and David B. Mustard (2003), “The Effects of Merit-
Based Financial Aid on Course Enrollment, Withdrawal and Completion in College” IZA
Discussion Paper.
20
[12] Dearden, Lorraine, Carl Emmerson, Christine Frayne, Costas Meghir, Karl Ashworth, Jay
Hardman, Yvette Hartfree, Sue Maguire, Sue Middleton and Debbi Smith (2002), “Education
Maintenance Allowance: The First Two Years A Quantitative Evaluation”, UK Department
for Education and Skills Research Report No.352.
[13] De Paola, Maria, Vincenzo Scoppa and Rosanna Nistico (2012), ”Monetary Incentives and
Student Achievement in a Depressed Labor Market: Results from a Randomized Experi-
ment,” Journal of Human Capital, University of Chicago Press, Vol. 6, No.1, pp 56 - 85.
[14] Dynarski, Susan (2002). “The Behavioral and Distributional Implications of Aid for Col-
lege.”American Economic ReviewVol. 92, No. 2, pp 279-285.
[15] Dynarski, Susan (2003). “Does Aid Matter? Measuring the Effect of Student Aid on College
Attendance and Completion.” American Economic Review Vol. 93, No. 1, pp 279- 288.
[16] Dynarski, Susan (2005). “Building the Stock of College-Educated Labor.” Working paper,
Kennedy School, Harvard University.
[17] Garibaldi Pietro, Francesco Giavazzi, Andrea Ichino and Enrico Rettore (2012), “College
cost and time to complete a degree: Evidence from tuition discontinuities”, The Review of
Economics and Statistics, Vol. 94, No.3, pp 699-711.
[18] Groen, Jeffrey, George Jakubson, Ronald G. Ehrenberg, Scott Condie and Albert Yung-
Hsu Liu (2006), “Program Design and Student Outcomes in Graduate Education”, NBER
Working Paper No. 12064.
[19] Hahn, J., Todd P. and van der Klaauw W. (2001), “Identification and Estimation of Treatment
Effects with a Regression-Discontinuity Design”, Econometrica Vol.69, No. 1, pp. 201-209.
[20] Hakkinen, I. and Uusitalo R. (2003), “The effect of a student aid reform on graduation: a
duration analysis”. Journal of Applied Econometrics, 5(1): 1-28.
[21] Hensen, Kari A., and Mack Shelley. 2003. “The Impact of Supplemental Instruction: Results
from a large, public, Midwestern University.” Journal of College Student Development, Vol.
44, No. 2, pp 250–59.
[22] Hoffer, Thomas B. and Vincent Welch (2006), “Time to Degree of U.S. Research Doctorate
Recipients”, National Science Foundation, Directorate for Social, Behavioral, and Economic
Sciences InfoBrief No. 312.
21
[23] Imbens, Guido W. and Joshua D. Angrist (1994), “Identification and Estimation of Local
Average Treatment Effects”, Econometrica, Vol. 62, No. 2, pp. 467-75.
[24] Kane, Thomas J. (2003), “A Quasi-Experimental Estimate of the Impact of Financial Aid on
College-Going”, NBER Working Paper No. 9703.
[25] Lavy, V.,Silva, O., Weinhardt F. (2009), “The Good, the Bad and the Average: Evidence on
the Scale and Nature of Ability Peer Effects in Schools”, NBER Working Paper 15600.
[26] Lee, David S. (2006), “Randomized Experiments from Non-random Selection in U.S. House
Elections”, Journal of Econometrics,Vol 142, No. 2, pp 675-697.
[27] Leslie, Larry L. and Paul T. Brinkman (1987), “Student Price Response in Higher Education:
The Student Demand Studies”, Journal of Higher Education, Vol. 58, No. 2, pp. 181-204.
[28] Leuven, Edwin, Hessel Oosterbeek and Bas van der Klaauw (2006), “The effect of financial
rewards on students’ achievement: Evidence from a randomized experiment”, Journal of the
European Economic Association Vol 8, No 6, pp 1243–1265.
[29] McCrary, Justin (2007), “Manipulation of the running variable in the regression discontinuity
design: a density test”, NBER Technical Working Paper No. 334.
[30] McPherson, Michael, and M. Schapiro (2006), “College Access: Opportunity or Privilege?”,
College Board.
[31] Mealli, Fabrizia and Carla Rampichini (2012),“Evaluating the Effects of University Grants
by using Regression Discontinuity Designs” Journal of Royal Statistic Society, Vol 175, No.
3, pp 775–798.
[32] OECD (2013), “Education at Glance”, Paris.
[33] Sacerdote, Bruce (2001) “Peer Effects with Random Assignment: Results for Dartmouth
Roommates”, Quarterly Journal of Economics Vol.116, No.2, pp 681-704.
[34] Schizzerotto, Antonio, Loris Vergolini and Nadir Zanini (2012), “Merit-Based Financial Aid
to Students From Low-Income Families and its Effects on Academic Performance”, Research
Institute for the Evaluation of Public Policies – IRVAP.
[35] van der Klaauw, W., (2002), “Estimating the Effect of Financial Aid Offers on College En-
rollment:a Regression-Discontinuity Approach”, International Economic Review, Vol. 43, No.
4., pp. 1249-1287.
22
[36] Van Ours, J. C. and Ridder G. (2003).”Fast Track or Failure: A Study of the Graduation
and Dropout Rates of Ph.D. Students in Economics”, Economic of Education Review, 22(2):
157-166
23
8 Appendix
Table 3: Descriptive Statistics
Variables Obs Mean Std. Dev. Min Max
Female 2430 0.6 0.48 0 1Age 2430 21.44 3.53 18 53
RegionCenter 2430 0.01 0.11 0 1
Center - North 2430 0.11 0.32 0 1Center - South 2430 0.04 0.21 0 1
Foreign 2430 0.06 0.24 0 1Islands 2430 0.09 0.29 0 1North 2430 0.37 0.48 0 1South 2430 0.28 0.45 0 1
High School Grade 2301 81.66 13.48 60 100Credits 2425 30.88 16.16 0 130
Average 1750 26 2.83 18 30
Scholarship typeIn sede 2430 0.4 0.49 0 1
Pendolari 2430 0.12 0.32 0 1Fuori Sede 2430 0.47 0.49 0 1
Note. Statistics for the freshman who enrolled at University of Bologna in the 2008/2009 academic year.
24
Table 4: Grant Effects on Students’ Achievement
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
Panel A: Credits
Higher Scholarships’ Effect 9.646** 10.141** 15.320*** 18.852***(3.951) (4.286) (4.722) (5.356)
Constant 25.602*** 24.944*** 21.076*** 18.280***(3.224) (3.329) (3.743) (4.292)
Obs 312 535 754 865
Panel B: Average Grade
Higher Scholarships’ Effect 1.867** 1.792* 2.063** 1.930*(0.846) (0.940) (1.026) (1.159)
Constant 25.157*** 24.935*** 24.764*** 24.919***(0.709) (0.774) (0.846) (0.956)
Obs 232 392 557 641
Panel C: Performance
Higher Scholarships’ Effect 0.106** 0.090* 0.113** 0.127**(0.044) (0.047) (0.052) (0.061)
Constant 0.670*** 0.674*** 0.652*** 0.645***(0.038) (0.038) (0.043) (0.051)
Obs 232 392 557 641
Panel D: P(Credits’Requirements)
Higher Scholarships’ Effect 0.229** 0.262** 0.380*** 0.472***(0.100) (0.114) (0.129) (0.149)
Constant 0.631*** 0.603*** 0.519*** 0.415***(0.083) (0.094) (0.106) (0.121)
Obs 312 535 754 865
Note. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (4)
where Zi is the vector of outcomes variables (Credits, Average Grade, Performance, P(Credit Requirements));F (Yi) is a vector whose elements are two polynomials (one for each side of the threshold) in the absolutedifference between the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on theleft side of the threshold. Robust standard errors in parentheses.
25
Table 5: Grant Effects on “Fuori Corso” proxy
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
Panel D: P(60 Credits)
Higher Scholarships’ Effect 0.057 0.054 0.086 0.100(0.062) (0.069) (0.078) (0.091)
Constant 0.045 0.044 0.033 0.035(0.043) (0.045) (0.052) (0.062)
R-squared 0.006 0.003 0.011 0.013Obs 312 535 754 865
Note. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (5)
where Zi is the outcomes variable P(Regular Credit Requirements). Credit Requirements is a dummy takingvalue 1 for students who have completed, at least, 60 credits in the academic year. Robust standard errors inparentheses.
26
Table 6: First Threshold Effects
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
Panel A: Credits
First Threshold Effect (950e) 11.863** 11.955* 18.113*** 23.483***(5.639) (6.127) (6.884) (8.117)
Constant 23.245*** 22.843*** 18.198*** 15.648**(4.715) (4.787) (5.459) (6.411)
Obs 162 273 369 421
Panel B: Average Grade
First Threshold Effect (950e) 3.026*** 2.845** 2.637* 2.248(1.044) (1.255) (1.363) (1.533)
Constant 24.482*** 24.443*** 24.537*** 24.699***(0.826) (1.022) (1.117) (1.275)
Obs 124 204 280 320
Panel C: Performance
First Threshold Effect (950e) 0.186*** 0.154** 0.175** 0.191**(0.062) (0.047) (0.077) (0.092)
Constant 0.608*** 0.634*** 0.605*** 0.606***(0.053) (0.057) (0.064) (0.076)
Obs 124 204 280 320
Panel D: P(Credits’Requirements)
First Threshold Effect (950e) 0.203 0.324* 0.480*** 0.649***(0.143) (0.165) (0.184) (0.209)
Constant 0.652*** 0.521*** 0.521*** 0.255(0.119) (0.135) (0.148) (0.160)
Obs 162 273 369 421
Note. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (6)
where Zi is the vector of outcomes variables (Credits, Average Grade, Performance, P(Credit Requirements));F (Yi) is a vector whose elements are two polynomials (one for each side of the threshold) in the absolutedifference between the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on theleft side of the threshold. Robust standard errors in parentheses.
27
Table 7: Second Threshold Effect
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
Panel A: Credits
Second Threshold Effect (730e) 7.093 8.989 13.296** 15.193**(5.504) (6.130) (6.612) (7.249)
Constant 28.177*** 26.520*** 23.642*** 20.691***(4.305) (4.777) (5.260) (5.942)
Obs 150 262 385 444
Panel B: Grades’Average
Second Threshold Effect (730e) 0.448 0.530 1.169 1.323(1.370) (1.424) (1.520) (1.685)
Constant 26.020*** 25.471*** 25.018*** 25.193***(1.203) (1.221) (1.285) (1.390)
Obs 108 188 277 321
Panel C: Performance
Second Threshold Effect (730e) 0.000 0.011 0.036 0.047(0.056) (0.059) (0.063) (0.069)
Constant 0.753*** 0.724*** 0.711*** 0.695***(0.044) (0.047) (0.050) (0.054)
Obs 108 188 277 321
Panel D: P(Credits’Requirements)
Second Threshold Effect (730e) 0.248* 0.241 0.307* 0.323(0.140) (0.157) (0.179) (0.209)
Constant 0.610*** 0.650*** 0.602*** 0.561***(0.117) (0.129) (0.150) (0.177)
Obs 150 262 385 444
Note. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (7)
where Zi is the vector of outcomes variables (Credits, Average Grade, Performance, P(Credits’ Requirements));F (Yi) is a vector whose elements are two polynomials (one for each side of the threshold) in the absolutedifference between the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on theleft side of the threshold. Robust standard errors in parentheses.
28
Table 8: Female Estimates
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
(a) FemalesPanel A: Credits
Higher Scholarships’ Effect 5.186 2.876 6.019 5.217(4.65) (5.01) (5.46) (6.1)
Constant 30.845*** 33.134*** 29.954*** 30.872***(3.91) (4.00) -4.43 -4.99
Obs 206 340 472 537
Panel B: Average Grade
Higher Scholarships’ Effect 2.320** 2.156** 2.540** 1.73(0.94) (1.04) (1.14) (1.2)
Constant 25.189*** 25.100*** 24.943*** 25.777***(0.82) (0.89) (0.96) (0.97)
Obs 161 266 369 418
Panel C: Performance
Higher Scholarships’ Effect 0.107** 0.076 0.102* 0.086(0.05) (0.05) (0.06) (0.07)
Constant 0.666*** 0.689*** 0.666*** 0.683***(0.04) (0.05) (0.05) (0.06)
Obs 161 266 369 418
Panel D: P(Credits’Requirements)
Higher Scholarships’ Effect 0.132 0.13 0.24 0.249(0.12) (0.14) (0.16) (0.2)
Constant 0.746*** 0.750*** 0.639*** 0.607***(0.1) (0.11) (0.13) (0.16)
Obs 206 340 472 537
Note: The reported results on outcome variables are only for female scholarship recipients enrolled at Universityof Bologna. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (8)
where Zi is the vector of outcomes variables (Credits, Average Grade, Performance, P(Credits’ Requirements));F (Yi) is a vector whose elements are two polynomials (one for each side of the threshold) in the absolutedifference between the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on theleft side of the threshold. Robust standard errors in parentheses.
29
Table 9: Male Estimates
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
(a) MalesPanel A: Credits
Higher Scholarship Effect 17.630*** 19.713*** 28.452*** 35.753***(6.36) (7.01) (7.43) (8.16)
Constant 16.442*** 14.173*** 9.702** 3.427(4.46) (4.52) (4.86) (5.07)
Obs 106 195 282 328
Panel B: Average Grade
Higher Scholarship Effect 0.896 1.246 1.435 2.899(1.8) (1.91) (2.12) (2.71)
Constant 25.147*** 24.388*** 24.267*** 22.632***(1.47) (1.51) (1.72) (2.35)
Obs 71 126 188 223
Panel C: Performance
Higher Scholarship Effect 0.09 0.127 0.159 0.244*(0.08) (0.09) (0.1) (0.13)
Constant 0.691*** 0.632*** 0.615*** 0.532***(0.07) (0.07) (0.08) (0.1)
Obs 71 126 188 223
Panel D: P(Credits’Requirements)
Higher Scholarships’ Effect 0.386** 0.450** 0.592*** 0.751***(0.17) (0.18) (0.19) (0.19)
Constant 0.451*** 0.405*** 0.360** 0.197(0.15) (0.14) (0.16) (0.16)
Obs 106 195 282 328
Note: The reported results on outcome variables are for male scholarship recipients enrolled at University ofBologna. OLS estimates of equation:
Zi = α+ F (Yi) + γDi + εi (9)
where Zi is the vector of outcomes variables (Credits, Average Grade, Performance, P(Credits’ Requirements));F (Yi) is a vector whose elements are two polynomials (one for each side of the threshold) in the absolutedifference between the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on theleft side of the threshold. Robust standard errors in parentheses.
30
Table 10: Regressions on Pre-treatment variables
Local Linear Polynom. 2nd Polynom. 3rd Polynom. 4th± 500 ± 850 ± 1200 ± 1400
Panel A: SexHigher Scholarship Effect 0.032 0.006 0.001 0.163
(0.11) (0.13) (0.14) (0.17)Constant 0.545*** 0.568*** 0.557*** 0.469***
(0.08) (0.10) (0.11) (0.13)Obs 321 552 780 899
Panel B: AgeHigher Scholarship Effect 0.999 0.911 0.449 -0.412
(0.71) (0.81) (0.97) (1.15)Constant 21.441*** 21.497*** 21.785*** 22.494***
(0.47) (0.54) (0.65) (0.80)Obs 321 552 780 899
Panel C: High-School GradeHigher Scholarship Effect 4.699 6.291 6.313 8.413
(3.44) (3.88) (4.41) (5.17)Constant 80.935*** 79.648*** 78.287*** 76.720***
(2.57) (2.87) (3.29) (3.89)Obs 296 510 725 836
Panel D: ISPEHigher Scholarship Effect 1536.229 2029.465 1445.873 3657.112
(2607.57) (3033.52) (3415.19) (4043.05)Constant 10063.320*** 10864.235*** 11198.737*** 10595.645***
(1855.30) (2142.27) (2480.20) (2968.13)
Obs 321 552 780 899
Panel E: P(Fuori Sede)Higher Scholarship Effect 0.049 0.072 0.096 0.024
(0.11) (0.13) (0.15) (0.17)Constant 0.411*** 0.367*** 0.349*** 0.408***
(0.08) (0.10) (0.11) (0.13)
Obs 321 552 780 899
Note. OLS estimates of equation:
Vi = α+ F (Yi) + γDi + εi (10)
where Vi is the vector of pre-treatment variables (Sex, Age, High-School Grade, ISPE, P(Fuori Sede)); F (Yi)is a vector whose elements are two polynomials (one for each side of the threshold) in the absolute differencebetween the ISEE declaration and the thresholds ; Di is a dummy taking value 1 for students on the left sideof the threshold. Robust standard errors in parentheses.
31
Fig
ure
1:G
raphic
alA
nly
sis
ofL
oca
lA
vera
geT
reat
men
tE
ffec
t
Note
.T
he
centr
al
lin
eis
asp
lin
efo
urt
h-o
rder
pol
yn
omia
lin
fam
ily
inco
me
ind
icat
or,
fitt
edse
par
ate
lyon
each
sid
eof
the
aggr
egat
eth
resh
old
at
zero
(fam
ily
inco
me
isn
orm
aliz
edas
the
dis
tan
cefr
omth
eab
ove
orb
elow
thre
shol
din
sym
met
ric
inte
rvals
).T
he
late
ral
lin
esare
the
90%
con
fid
ence
inte
rval.
Sca
tter
poi
nts
are
aver
aged
over
100-
un
itin
terv
als.
Th
eC
red
its
outc
ome
isth
enu
mb
erof
cred
its
obta
ined
inth
efi
rst
year
ofen
rolm
ent.
Th
eA
vera
geis
the
aver
age
grad
eob
tain
edin
the
firs
tye
arof
enro
lmen
t.P
erfo
rman
ceis
aco
nst
ruct
edin
dex
of
aca
dem
icach
ieve
men
tth
at
com
bin
edb
oth
the
cred
its
and
the
grad
es’
aver
age
outc
omes
.P
(Sat
isfy
Cre
dit
s’R
equ
irem
ents
)is
the
pro
bab
ilit
yof
hav
ing
sati
sfy
the
requ
irem
ents
for
the
main
ten
ance
and
non
-rem
bu
rsem
ent
ofth
esc
hol
arsh
ipat
the
end
ofth
efi
rst
yea
r.
32
Fig
ure
2:E
vid
ence
onso
rtin
gan
dco
nti
nuit
yco
ndit
ions
Note
.T
he
centr
al
lin
eis
asp
lin
efo
urt
h-o
rder
pol
yn
omia
lin
fam
ily
inco
me
ind
icat
or,
fitt
edse
par
ate
lyon
each
sid
eof
the
aggr
egat
eth
resh
old
at
zero
(fam
ily
inco
me
isn
orm
aliz
edas
the
dis
tan
cefr
omth
eab
ove
orb
elow
thre
shol
din
sym
met
ric
inte
rvals
).T
he
late
ral
lin
esare
the
90%
con
fid
ence
inte
rval
.S
catt
erp
oints
are
aver
aged
over
100-
un
itin
terv
als.
Th
eS
exou
tcom
eis
the
pro
bab
ilit
yof
bei
ng
fem
ale
stu
den
ts.
Age
isth
est
ud
ents
’age
at
the
enro
lmen
tin
the
firs
tye
arof
stu
dy.
Hig
h-S
chool
Gra
de
isth
efi
nal
grad
ein
the
hig
h-s
chool
dip
lom
a.
ISP
Eis
anin
dic
ator
onfa
mil
yas
sets
.P
(”F
uori
Sed
e”)
isth
ep
rob
abil
ity
ofre
ceiv
ing
the
hig
hes
t”F
uor
iS
ede”
sch
ola
rsh
ip.
33
Figure 3: Test on Manipulation
Note.Scatter points of density of observations around the aggregate threshold. The lateral lines are the 95%confidence interval.
34